Updated project metadata. Using proteogenomics, we identified and analyzed 25,172 major histocompatibility complex class I-associated peptides (MAPs) isolated from B lymphocytes of 18 individuals who collectively expressed 27 HLA-A,B allotypes. While 58% of genes were the source of 1-64 MAPs per gene, 42% of genes were not represented in the immunopeptidome. Overall, we estimate the immunopeptidome presented by 27 HLA-A,B allotypes covered only 17% of exomic sequences expressed in subjects’ cells. We identified several features of transcripts and proteins that enhance MAP production. From these data we built a logistic regression model that predicts with high accuracy whether a gene from our dataset or from independent datasets would generate MAPs. Our results show preferential selection of MAPs from a limited repertoire of gene products with distinct features. The notion that the immune system can monitor MAPs covering only a fraction of the protein coding genome has profound implications in autoimmunity and cancer immunology.